Optimized Hybrid Rotation Invariant LBP fused Lightweight MobileNetV2 Model for Deep Visual Inspection of Apple Bruises

Abstract

Bruise detection in apples is a critical task which is faced in post-harvest quality control of the fruit. The bruises significantly impact consumer preference, shelf life, and overall market value of the fruit. Traditional inspection methods are labor-intensive, subjective and prone to inconsistencies. In order to cope up with huge number of production demands, the real necessity is felt in the development of efficient, automated, and non-destructive techniques classification of the fruit. In this study, we have proposed a optimized rotation invariant local binary pattern (LBP) model fused with the light weight MobileNetV2. LBP offers local texture feature of the fruit. The rotation invariant feature of LBP avoids the rotational effects, likely to occur at the time of image capturing of the samples. The performance of hybrid model has been assessed by several metrics. The result of assessment was estimated at 98% for classification accuracy and a 0.98 for AUC measure of ROC curve, the model's initial findings demonstrates its efficacy.